Abstract: With the continuous development of digital power systems, research on digital twin models of high-value power equipment has become a hot topic. In order to save computing resource, model reduction methods are usually used, among which proper orthogonal decomposition (POD) is one of the most commonly used methods. The traditional POD method cannot efficiently extract grid node information when constructing snapshots, resulting in a large amount of calculation and waste of storage resources. So, this paper proposes a snapshot partition POD method based on field gradient. The calculation domain is partitioned based on the field gradient of each node in the snapshot. By reducing the number of nodes in different partitions to various degrees, efficient POD calculation is achieved. The polarity reversal electric field of the bushing on the converter transformer valve side is used as an example to verify the feasibility and efficiency of the partition reduction method proposed in this paper. The research results show that, compared with the finite element method, the calculation time is shortened by 97.1%. Compared with the traditional POD method, the number of nodes in the snapshot is reduced by 95.2%, saving 91.7% of storage resources. At the same time, the accuracy of the partitioned reduced-order calculation is high, with an average error of only 0.80%, which realizes the efficient calculation of the polarity reversal electric field of the valve-side casing.